Repeated Games and Reinforcement Learning
Organizer
Speaker
Feng Fu
Time
Wednesday, July 24, 2024 2:00 PM - 3:30 PM
Venue
A3-2a-302
Online
Zoom 559 700 6085
(BIMSA)
Abstract
Very recently, the notion of cooperative AI has been advocated as one of the solutions to ensure beneficial AI technologies. In this light, reciprocal cooperation has been extensively studied using the Iterated Prisoner’s Dilemma (IPD) games. Despite the astronomically vast individual behavioral strategy space for IPD game interactions, the so-called Zero-Determinant (ZD) extortionate strategy is a set of rather simple memory-one strategies that can unilaterally set a linear relationship between itself and its opponent. This new finding of ZD strategies has greatly spurred new waves of work from diverse fields, such as network science, computer science, and applied mathematics, aiming to shed light on the robustness and resilience of cooperation through the natural selection of IPD strategies. However, one open issue remains to be fully addressed: can extortionate ZD strategies be outperformed at all, even in simple head-to-head IPD matches? We will use machine learning and analytical tools to search for strategies that are unyielding to opponents’ extortion (i.e., strategies that can make extortion backfire and even outperform ZD under certain conditions). This will bring a new perspective by combining game theory and AI techniques for optimizing winning strategies in non-zero-sum games.
Speaker Intro
Feng Fu is currently an associate professor of applied mathematics in the Department of Mathematics and holds an adjunct appointment in the Department of Biomedical Data Science at the Geisel School of Medicine, Dartmouth. Before joining Dartmouth in September 2015, he was a senior postdoc in theoretical biology under Sebastian Bonhoeffer at ETH Zurich starting in September 2012. Prior to that, he did his dissertation research and a subsequent postdoc training with Martin Nowak and Nicholas Christakis at Harvard University from 2007 to 2012. He is interested in evolutionary game theory with applications to real-world problems, including reinforcement learning dynamics, neuroscience, and behavioral epidemiology. He received his B.Sc. in Theoretical and Applied Mechanics from Fudan University in 2004 and his Ph.D. in Dynamics and Control from Peking University in 2010. He was the recipient of Dartmouth's Dean of the Faculty Award for Outstanding Mentoring and Advising in 2021.